AI news, Tuesday, June 9, 2026

TLDR

• Google pays SpaceX $920M/month for GPUGraphics processing unitA highly parallel processor that does the large matrix math behind training and running models.NVIDIA ↗ compute (Jun 5) — $30B deal through mid-2029; SpaceX is now effectively an AI cloud provider after absorbing xAI's compute fleet in the Feb merger.
• PhysicsX raises $300M Series C at $2.4B (Jun 8–9) — Temasek-led round for Large Physics Models in aerospace, chip design, and manufacturing — the week's biggest industrial AI signal.
• Linux Foundation launches Tokenomics Foundation (Jun 8–10) — New standards body to bring FinOps discipline to AI tokenA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗ spending, backed by Google Cloud, Microsoft, IBM, and JPMorganChase; timed to FinOps X in San Diego this week.
• AI token budgets breaking in production (Jun 5) — Some companies exhausted their entire 2026 AI coding budget by April; TechCrunch deep-dive finds enterprise AI spend is the fastest-growing and least-governed line item in IT.
• OpenCode v1.16.0 (Jun 5) — Major update to the 160K-star open-source coding agent: managed workspace cloning, Copilot credit tracking, skill discovery, and session replay.

---

Action Items

Things your engineering team should look at or do this week.

• Audit your 2026 AI token spend now — TechCrunch's June 5 investigation found enterprises routinely 3× over annual AI token budgets by Q2. Pull actuals from your cloud billing dashboards (AWS Bedrock, Azure AI Foundry, Anthropic console) before the next billing cycle. The Tokenomics Foundation is launching open specs at FinOps X this week — bookmark their site for the cost-governance framework. (TechCrunch | Linux Foundation)

• Evaluate OpenCode v1.16.0 if you're on GitHub Copilot — The new release adds per-session Copilot AI Credit tracking so you can see token drain from agenticDescribes AI systems where a model uses tools and decides its own steps instead of following a fixed script.Anthropic ↗ sessions in real time — directly useful now that Copilot switched to usage-based billing on June 1. (GitHub releases | opencode.ai)

• Watch for MiniMax M3 open weightsA model whose trained weights are published for anyone to download and run.NTIA ↗ ~June 10–11 — The API has been live since June 1 (covered Jun 6 digest); the open weights and technical report are expected on Hugging Face this week. If your team needs an open-weight model for coding or long-context tasks, this is worth evaluating as an alternative to proprietary APIs. Note: benchmarks (59% SWE-BenchA benchmark that tests whether a model can resolve real GitHub issues by editing code in popular Python repositories.Jimenez et al. 2023 ↗ Pro) are self-reported and unverified by independent labs. (MiniMax Blog | MarkTechPost)

---

Models & Tools

• OpenCode v1.16.0 (Jun 5) Open Source Agents — The open-source terminal coding agent (160K+ GitHub stars, 7.5M MAU) released its latest update: managed workspace cloning that preserves dirty/untracked files across sessions, skill discovery and file-based agent loading, GitHub Copilot AI Credit usage tracking for the new token-based billing model, --replay flag for interactive session debugging, and improved startup time. Model-agnostic across 75+ providers with LSP diagnostics feed-back. (GitHub | opencode.ai)

---

APIs & Pricing

• Google-SpaceX: $30B compute deal reveals AI infrastructure pressure (Jun 5) Infra Trending — Bloomberg reported Google will pay SpaceX $920M/month from October 2026 through June 2029 for ~110,000 NVIDIA GPUs. Google's internal note described it as "bridge capacity" to meet "surging demand" for Gemini Enterprise that exceeded internal forecasts. SpaceX became a de-facto GPU cloud provider after absorbing xAI's Colossus cluster in February. The scale of the deal is a signal of how constrained frontier inferenceRunning a trained model on new input to get an output, as opposed to training it.NVIDIA ↗ capacity remains even for hyperscalers. (Bloomberg | TechCrunch)

---

Industry & Business

• PhysicsX raises $300M Series C, valued at $2.4B (Jun 8–9) Robotics/IoT Infra — London-based startup building AI-native simulation for industrial engineering closed an oversubscribed round led by Temasek, with NVIDIA, Siemens, Applied Materials, and General Catalyst participating. PhysicsX doubled YoY revenue and tripled booked revenue, with customers in aerospace, semiconductor design, defense, and automotive. The capital funds frontier research into Large Physics Models (LPMs) — essentially foundation models for physical simulation — a category that's getting serious traction alongside large language and vision models. (PhysicsX Newsroom | TNGlobal | The Next Web)

• 2026 AI-attributed layoffs: 55% of all tech cuts cite AI (Jun 8 data) Hiring — Through June 8, 247 layoff events have affected 184K workers; 55% explicitly cite AI or automation, covering 152K people. The split: junior developer and content roles contracting sharply; ML infrastructure, AI safety, and MLOps roles remain in strong demand with companies like Atlassian cutting 1,600 general roles while hiring 800 AI-focused positions in the same announcement. (allwork.space | CBS News)

---

AI in Production

• Enterprise AI token budgets collapsing mid-year (Jun 5) Production Infra — TechCrunch's investigation found Uber burned through its entire 2026 AI coding budget by April; Microsoft revoked developer Claude Code licenses due to cost; Priceline reported renewal costs 4–5× higher than expected. The root cause: agentic workflows that chain model calls consume tokens 10–30× faster than chat usage patterns, but most budgets were modeled on chat rates. Mitigation now in play: model routing (sending simpler subtasks to cheaper models), output caching, per-user spend caps, and prompt-length governance. The Tokenomics Foundation's open specs (launching this week) are positioned as the industry's FinOps answer. (TechCrunch | Linux Foundation press release)